影像處理技術應用於數位單眼相機手震偵測

摘要本論文的目的在於提出一個新想法來解決數位單眼相機手震的問題，我們主要使用數位影像處理來分辨出照片是否因手震而模糊，並希望即使手震程度非常小也能夠分辨出來。在本論文我們提出的方法分為兩部分，第一部份是用數位影像處理去做數位相機的手震偵測。第二部份是屬於實驗結果。在第一部份我們拍了兩張照片，一張是清晰的一張是模糊的，我們是用Sobel邊界偵測的方法來區分兩張照片的不同。接著我們把兩張照片都做擴張與侵蝕來減少一些我們不想要的邊界，最後我們使用骨架化來使得邊界變得更清楚以便顯示出兩張照片不同。第二部份我們先拍攝兩種照片，一種是在不同條件下拍攝同一個物體，另一個是在同一個條件下拍攝不同物體，並將兩種照片都經過第一部份的方法做影像處理。最後我們拿這些照片去讓人們挑選看結果我們的方法的效率是否比較高。本論文的貢獻如下: (1) 提高初學者的辨識速度: 對於單眼數位相機的初學者而言要辨別拍出的相片是否模糊是很困難的。這個方法可以幫助初學者來辨識相片是不是模糊的。 (2) 減少錯過拍攝的最佳時機的機率: 使用此方法使用者快速的分辨照片是不是模糊的並不需要放大照片來檢視，換句話說這方法可以減少在拍攝會動的物體時，因為要檢視照片而錯過最佳時機的機率。 (3) 避免拍出輕微模糊的照片: 有些模糊的照片在單眼數位相機的液晶螢幕上是不容易分辨的，但使用本方法可以在相機的液晶螢幕上清楚的辨識出相片是否模糊並且避免拍出輕微模糊的照片。 (4) 增加辨識相片是否模糊的準確率: 透過本研究的模擬結果，影像處理前與影像處理後的準確率從66.67%提升到79.17%。

關鍵字

數位影像處理；數位單眼相機

並列摘要

Abstract In this thesis, we propose a new way to detect shake on DSLR camera. Our main propose in this thesis is to use digital image processing to discriminate the blurred images caused by shake, even the shake is very light that is difficult to detect on the DSLR camera's LCD screen. In this thesis, the methods we propose have two parts. One is applying image processing to make camera vibration detection. The other part is the experimental results. On the first part, we shoot two photos, one is clear and another is blurred to detect the difference after digital image processing between them. We use the Sobel filter to find the edges of the two images. After that, we use dilation and erosion to these two images filter to reduce the edges we don’t want. Last we use skeletonization and pruning to make the edges become clear to display the difference between clear photo and blurred photo. On the second part, we will use the purposed methods on the first part to do same experiments. We take two kinds of photos: one kind is to shoot blurred photos of the same objects in different conditions, and another kind is to shoot blurred photos of different objects in the same conditions. Finally, we take these photos to be processed according to our proposed methods to see if they display the blurred edges to prove the effectness of our methods, what photos are clear, what photos are blurred. In this thesis, the contributions of our research are as follow: (1) Promote beginner’s discriminate speed: It is not easy for beginners to discriminate if photos blurred or not. This method can help the beginners to discriminate the blurred photos. (2)Reduce the probability of missing perfect shooting time: Using this method, one can discriminate blurred photo or clear photos quickly. Hence you don’t need to zoom in to check if photos blurred or not. In other words, this method can reduce the probability of missing perfect shooting time when you shot moving objects. (3) Prevent shooting light blurred photos: Light blurred photos are not obvious on camera’s LCD screen. By using our methods, these light blurred photos will very obvious on camera’s LCD screen, so you can delete these photos and prevent shooting light blurred photos. (4)Add the accuracy of discriminate photos: Form simulation results, it tells us that the accuracy of discrimination of blurred photos increases from 66.67% to 79.17%.